“Whatever gets published mustn't sound like a machine.” A client told me that when we started working together. I wrote it down, because it sums up the job: AI writes fast and nobody wants to read a machine.
I'm Nico, a freelance digital marketing consultant in Seville. I use AI every day, in my own marketing and my clients', as part of digital marketing. Here's what I give it and what I never give it. And how I review before anything goes out with my name or yours on it.
What can AI do for a small business's marketing?
Whatever repeats and has clear rules. A first draft based on your sources and your style guide. Sorting Search Console and Analytics data into a spreadsheet. Preparing images and cover images. The mechanical steps of a report or a maintenance job. It does all that well and fast, as long as someone checks it afterwards.
| Task | What AI does | What I do | Time |
|---|---|---|---|
| Blog article | Draft with my sources, my own data and my voice guide | Edit, decide, publish | From 2.5 h to 40 min |
| Monthly report | Gathering data and putting together the spreadsheet and presentation | The interpretation and the three decisions, with the client | Putting it together, next to nothing; the interpretation, all of it |
| Proposals | Structure and draft based on the meeting | Price, scope and who it's not for | Less time putting it together |
| Website maintenance | Backup, update, check and alert, with my own scripts | The issues the system can't fix | 30 minutes per website per week |
| Images | Icons and cover images from a brief | Choose, crop, discard | Less preparation time |
It's structural support, to simplify processes. None of those processes runs without me. When someone sells me “one hundred per cent automated marketing”, I know what's going to happen. Content that sounds like a machine and nobody to answer when something breaks.
What I never delegate
Five things, and they're not negotiable.
- What gets published and what doesn't. AI suggests topics; the filter is mine: if a model could write it without leaving the house, it doesn't go out.
- Figures. Every number carries a source or my own dated data. If AI gives me a figure without a source, it gets flagged and looked up. If it can't be found, it comes out.
- Names and web addresses. A made-up client name or URL does more damage than a weak paragraph.
- The voice. I keep a list of phrases that AI repeats and I don't say. Just one in the text and the sentence gets rewritten.
- The button. Nothing publishes itself. The system leaves drafts; publishing is a human act, with a read-aloud first.
What Google says (and what Ahrefs measured)
Google has it in writing in its guidance on AI-generated content. When creating content, focus on accuracy, quality and relevance, “especially when it's generated automatically”. John Mueller said it in the June 2024 office hours, about AI drafts reviewed by editors. “What we care about is the overall quality of what you end up publishing.”
What it does go after is in its spam policies, in the section on scaled content. “Using generative AI tools or other similar tools to generate many pages without adding value for users.” Since 15 May 2026 those policies also apply to AI responses in Search itself. And between 18 and 21 August 2026 Google rolled out a global spam update, according to its status dashboard.
Ahrefs put numbers on it on 27 July 2026, with 331,000 pages analysed. Pages with little AI-generated content got between two and three times more impressions than those that were almost entirely AI-generated. The tool works fine. What fails is releasing whatever comes out without touching it.
And the Spanish context, according to Spain's INE. In the first quarter of 2025, AI technologies were used by 21.1% of companies with ten or more employees. Among those with fewer than ten, 13.4%. Almost nine in ten small businesses still don't use it. Those that use it with good judgement get ahead; those that use it without reviewing run a risk.
How I review before publishing
The same process for every piece, whether it takes ten minutes or forty:
- Value filter. Could a model write it without leaving the house? If so, back to the start with a data point of my own in it.
- Figures and names. Every number with its linked source and its date. Every name and every web address, checked.
- Voice pass. Template formulas, symmetrical lists of three, long sentences, words I don't use. Just one and it gets rewritten.
- Reading aloud. If I wouldn't say it in a meeting with the manager of a brokerage, it goes.
- Three attempts. If the text hasn't passed by the third review, it stays as a draft and gets shelved. Nothing goes out just because I'm tired of it.
What I see with my clients
What worries people most in the first meeting is the line I started with: that it'll sound like a machine. I get it. I've read company blogs written with unreviewed AI and you can spot them a mile off. The same openings, the same lists of three, no in-house data.
My answer is the system above. On the blogs I run, around 40 articles have gone out with AI in the draft and my review on every one. The time per article dropped from two and a half hours to 40 minutes. That time doesn't disappear: it goes into getting to know the business and deciding, which is what AI can't do. I explain how that blog is planned in Blogging for your small business: what to write and how often to post.
If you want to use it in your business
Using AI for your marketing and not sure whether what comes out helps or hurts? In 30 minutes we review a piece together using the checklist above. Free Diagnostic Session, two slots a week. Book yours. And if you'd rather learn to set up the system yourself, that's covered in digital marketing training and mentoring.
Sources
- Google Search Central, Guidance on using generative AI content on your website (“focus on accuracy, quality and relevance, especially when it's generated automatically”): https://developers.google.com/search/docs/fundamentals/using-gen-ai-content?hl=es
- Google Search Central, Spam policies for Google web search (scaled content abuse): https://developers.google.com/search/docs/essentials/spam-policies?hl=es
- Google Search Central, documentation changelog (15 May 2026: spam policies apply to generative AI responses): https://developers.google.com/search/updates
- Google Search Status Dashboard, August 2026 spam update (18 to 21 August, global): https://status.search.google.com/incidents/LEubPCm2octf2uMqCFKE
- Google Search Central, SEO office hours, June 2024 (John Mueller: “what we care about is the overall quality of what you end up publishing”): https://developers.google.com/search/help/office-hours/2024/june
- Ahrefs, Google doesn't punish AI content; it punishes bad content (331,000 pages; 27 July 2026): https://ahrefs.com/blog/google-doesnt-punish-ai-content/
- INE, Survey on ICT usage and e-commerce in companies, 2024 – first quarter of 2025 (21.1% and 13.4% of companies using AI technologies; 22 October 2025): https://www.ine.es/dyngs/Prensa/ETICCE20241T2025.htm
- Own data: log of the Auve Nico writing and review system (2026) and client quotes gathered in the 2026 internal audit (anonymised).
Freelance digital marketing consultant in Seville. I work on the digital presence of professionals and small businesses with one simple principle: your website as your operations hub. Building websites, doing SEO and running social media with my own hands since 2018.
Get to know me
